The answer to this question was a little more involved for me.
I found no discernible difference between putting qunit in a package, and just including qunit sources in my /client files. My difficulty was that sometimes tests seemed to run, sometimes not at all, and frequently a mysterious "global error" would appear in my test results.
This was called by qunit attempting to automatically launch the test run before my own code had loaded tests. I found no good solution to prevent the automatic behavior. My eventual solution was to let qunit finish its (empty) automatic test run, and then to call Qunit.init(), load tests, then Qunit.start().